A robust real-time surface reconstruction method on point clouds captured from a 3D surface photogrammetry system.

Area of Science:

• Computer Vision and Graphics
• Medical Imaging and Image Processing
• Computational Geometry

Background:

• Traditional surface reconstruction methods often rely on solving complex partial differential equations, leading to slow processing times.
• 3D surface photogrammetry systems generate point clouds that require efficient and accurate reconstruction for various applications.
• Variational approaches, while robust, are computationally intensive and not suitable for real-time reconstruction needs.

Purpose of the Study:

• To develop a robust and real-time surface reconstruction method for point clouds acquired via 3D surface photogrammetry.
• To improve the speed and accuracy of surface reconstruction without the computational burden of traditional methods.

Main Methods:

• Developed a novel surface reconstruction method that avoids solving partial differential equations.
• Utilized sparse regression (SR) to approximate point cloud data by leveraging the overcomplete nature of acquired point clouds.
• Introduced a modified sparse regression (MSR) model to handle varying noise levels and occlusions by incorporating a Laplacian prior.

Main Results:

• Achieved sub-millimeter reconstruction accuracy on clinical point clouds using both SR and MSR models.
• Reduced reconstruction time by two orders of magnitude, achieving sub-second processing.
• Demonstrated robust and consistent performance with the MSR model on point clouds containing inconsistent occlusions.

Conclusions:

• The developed method offers a fast and robust solution for surface reconstruction from 3D photogrammetry point clouds.
• Sub-millimeter accuracy and sub-second reconstruction time make the method suitable for real-time applications.
• The technique is particularly promising for real-time motion tracking in radiotherapy, enhancing quantification through clear surface structures.